CS6220: DATA MINING TECHNIQUES

Transcription

1 CS6220: DATA MINING TECHNIQUES Matrix Data: Classification: Part 2 Instructor: Yizhou Sun yzsun@ccs.neu.edu September 21, 2014

2 Methods to Learn Matrix Data Set Data Sequence Data Time Series Graph & Network Classification Decision Tree; Naïve Bayes; Logistic Regression SVM; knn HMM Label Propagation Clustering Frequent Pattern Mining K-means; hierarchical clustering; DBSCAN; Mixture Models; kernel k-means Apriori; FP-growth GSP; PrefixSpan Prediction Linear Regression Autoregression SCAN; Spectral Clustering Similarity Search Ranking DTW P-PageRank PageRank 2

3 Matrix Data: Classification: Part 2 Bayesian Learning Naïve Bayes Bayesian Belief Network Logistic Regression Summary 3

4 Bayesian Classification: Why? A statistical classifier: performs probabilistic prediction, i.e., predicts class membership probabilities Foundation: Based on Bayes Theorem. Performance: A simple Bayesian classifier, naïve Bayesian classifier, has comparable performance with decision tree and selected neural network classifiers Incremental: Each training example can incrementally increase/decrease the probability that a hypothesis is correct prior knowledge can be combined with observed data Standard: Even when Bayesian methods are computationally intractable, they can provide a standard of optimal decision making against which other methods can be measured 4

5 Basic Probability Review Have two dices h 1 and h 2 The probability of rolling an i given die h 1 is denoted P(i h 1 ). This is a conditional probability Pick a die at random with probability P(h j ), j=1 or 2. The probability for picking die h j and rolling an i with it is called joint probability and is P(i, h j )=P(h j )P(i h j ). If we know P(i h j ), then the so-called marginal probability P(i) can be computed as: P i = j P(i, h j ) For any X and Y, P(X,Y)=P(X Y)P(Y) 5

6 Bayes Theorem: Basics Bayes Theorem: P( h X) P( X h) P( h) P( X) Let X be a data sample ( evidence ) Let h be a hypothesis that X belongs to class C P(h) (prior probability): the initial probability E.g., X will buy computer, regardless of age, income, P(X h) (likelihood): the probability of observing the sample X, given that the hypothesis holds E.g., Given that X will buy computer, the prob. that X is , medium income P(X): marginal probability that sample data is observed P X = h P X h P(h) P(h X), (i.e., posterior probability): the probability that the hypothesis holds given the observed data sample X 6

7 Classification: Choosing Hypotheses Maximum Likelihood (maximize the likelihood): h ML arg max P( DX h) h H Maximum a posteriori (maximize the posterior): Useful observation: it does not depend on the denominator P(X) h MAP arg max P( h h H DX ) arg max P( DX h) P( h) h H 7

8 Classification by Maximum A Posteriori Let D be a training set of tuples and their associated class labels, and each tuple is represented by an p-d attribute vector X = (x 1, x 2,, x p ) Suppose there are m classes Y {C 1, C 2,, C m } Classification is to derive the maximum posteriori, i.e., the maximal P(Y=C j X) P( X Y C ) This can be derived from Bayes theorem P( Y C X) j j P( Y C ) j P( X) Since P(X) is constant for all classes, only needs to be maximized P( y, X) P( X y) P( y) 8

9 Example: Cancer Diagnosis A patient takes a lab test with two possible results (+ve, -ve), and the result comes back positive. It is known that the test returns a correct positive result in only 98% of the cases (true positive); a correct negative result in only 97% of the cases (true negative). Furthermore, only of the entire population has this disease. 1. What is the probability that this patient has cancer? 2. What is the probability that he does not have cancer? 3. What is the diagnosis? 9

10 Solution P(cancer) =.008 P( cancer) =.992 P(+ve cancer) =.98 P(-ve cancer) =.02 P(+ve cancer) =.03 P(-ve cancer) =.97 Using Bayes Formula: P(cancer +ve) = P(+ve cancer)xp(cancer) / P(+ve) = 0.98 x 0.008/ P(+ve) = / P(+ve) P( cancer +ve) = P(+ve cancer)xp( cancer) / P(+ve) = 0.03 x 0.992/P(+ve) =.0298 / P(+ve) So, the patient most likely does not have cancer. 10

11 Matrix Data: Classification: Part 2 Bayesian Learning Naïve Bayes Bayesian Belief Network Logistic Regression Summary 11

12 Naïve Bayes Classifier Let D be a training set of tuples and their associated class labels, and each tuple is represented by an p-d attribute vector X = (x 1, x 2,, x p ) Suppose there are m classes Y {C 1, C 2,, C m } Goal: Find Y max P Y X = P(Y, X)/P(X) P X Y P(Y) A simplified assumption: attributes are conditionally independent given the class (class conditional independency): p P( X C j) P( x k 1 k C j) P( x 1 C j) P( x 2 C j)... P( x p C j) 12

13 Estimate Parameters by MLE Given a dataset D = {(X i, Y i )}, the goal is to Find the best estimators P(C j ) and P(X k = x k C j ), for every j = 1,, m and k = 1,, p that maximizes the likelihood of observing D: L = = i i ( P X i, Y i = Estimators of Parameters: k P X ik Y i )P(Y i ) i P X i Y i P(Y i ) P C j = C j,d / D ( C j,d = # of tuples of C j in D) (why?) P X k = x k C j : X k can be either discrete or numerical 13

14 Discrete and Continuous Attributes If X k is discrete, with V possible values P(x k C j ) is the # of tuples in C j having value x k for X k divided by C j, D If X k is continuous, with observations of real values P(x k C j ) is usually computed based on Gaussian distribution with a mean μ and standard deviation σ Estimate (μ, σ 2 ) according to the observed X in the category of C j Sample mean and sample variance P(x k C j ) is then P( X k xk C j) g( xk, C i, C ) i Gaussian density function 14

15 Naïve Bayes Classifier: Training Dataset Class: C1:buys_computer = yes C2:buys_computer = no Data to be classified: X = (age <=30, Income = medium, Student = yes Credit_rating = Fair) age income studentcredit_rating_comp <=30 high no fair no <=30 high no excellent no high no fair yes >40 medium no fair yes >40 low yes fair yes >40 low yes excellent no low yes excellent yes <=30 medium no fair no <=30 low yes fair yes >40 medium yes fair yes <=30 medium yes excellent yes medium no excellent yes high yes fair yes >40 medium no excellent no 15

16 Naïve Bayes Classifier: An Example age income studentcredit_rating_comp <=30 high no fair no <=30 high no excellent no high no fair yes >40 medium no fair yes >40 low yes fair yes >40 low yes excellent no low yes excellent yes <=30 medium no fair no <=30 low yes fair yes >40 medium yes fair yes <=30 medium yes excellent yes medium no excellent yes high yes fair yes >40 medium no excellent no P(C i ): P(buys_computer = yes ) = 9/14 = P(buys_computer = no ) = 5/14= Compute P(X C i ) for each class P(age = <=30 buys_computer = yes ) = 2/9 = P(age = <= 30 buys_computer = no ) = 3/5 = 0.6 P(income = medium buys_computer = yes ) = 4/9 = P(income = medium buys_computer = no ) = 2/5 = 0.4 P(student = yes buys_computer = yes) = 6/9 = P(student = yes buys_computer = no ) = 1/5 = 0.2 P(credit_rating = fair buys_computer = yes ) = 6/9 = P(credit_rating = fair buys_computer = no ) = 2/5 = 0.4 X = (age <= 30, income = medium, student = yes, credit_rating = fair) P(X C i ) : P(X buys_computer = yes ) = x x x = P(X buys_computer = no ) = 0.6 x 0.4 x 0.2 x 0.4 = P(X C i )*P(C i ) : P(X buys_computer = yes ) * P(buys_computer = yes ) = P(X buys_computer = no ) * P(buys_computer = no ) = Therefore, X belongs to class ( buys_computer = yes ) 16

17 Avoiding the Zero-Probability Problem Naïve Bayesian prediction requires each conditional prob. be non-zero. Otherwise, the predicted prob. will be zero p P( X C j) P( xk C j) k 1 Use Laplacian correction (or Laplacian smoothing) Adding 1 to each case P x k = v C j = n jk,v+1 C j,d +V where n jk,v is # of tuples in C j having value x k = v, V is the total number of values that can be taken Ex. Suppose a training dataset with 1000 tuples, for category buys_computer = yes, income=low (0), income= medium (990), and income = high (10) Prob(income = low buys_computer = yes ) = 1/1003 Prob(income = medium buys_computer = yes ) = 991/1003 Prob(income = high buys_computer = yes ) = 11/1003 The corrected prob. estimates are close to their uncorrected counterparts 17

18 *Smoothing and Prior on Attribute Distribution Discrete distribution: X k C j ~ θ P X k = v C j, θ = θ v Put prior to θ In discrete case, the prior can be chosen as symmetric Dirichlet distribution: θ~dir α, i. e., P θ v θ v α 1 posterior distribution: P θ X 1k,, X nk, C j P X 1k,, X nk C j, θ P θ, another Dirichlet distribution, with new parameter (α + c 1,, α + c v,, α + c V ) c v is the number of observations taking value v Inference: P X k = v X 1k,, X nk, C j = P(X k = v θ)p θ X 1k,, X nk, C j dθ = c v + α c v + Vα Equivalent to adding α to each observation value v 18

19 *Notes on Parameter Learning Why the probability of P X k C j estimated in this way? Fall06/reading/NB.pdf is 19

20 Naïve Bayes Classifier: Comments Advantages Easy to implement Good results obtained in most of the cases Disadvantages Assumption: class conditional independence, therefore loss of accuracy Practically, dependencies exist among variables E.g., hospitals: patients: Profile: age, family history, etc. Symptoms: fever, cough etc., Disease: lung cancer, diabetes, etc. Dependencies among these cannot be modeled by Naïve Bayes Classifier How to deal with these dependencies? Bayesian Belief Networks 20

21 Matrix Data: Classification: Part 2 Bayesian Learning Naïve Bayes Bayesian Belief Network Logistic Regression Summary 21

22 Bayesian Belief Networks (BNs) Bayesian belief network (also known as Bayesian network, probabilistic network): allows class conditional independencies between subsets of variables Two components: (1) A directed acyclic graph (called a structure) and (2) a set of conditional probability tables (CPTs) A (directed acyclic) graphical model of causal influence relationships Represents dependency among the variables Gives a specification of joint probability distribution Nodes: random variables X Z Y P Links: dependency X and Y are the parents of Z, and Y is the parent of P No dependency between Z and P conditional on Y Has no cycles 22 22

23 A Bayesian Network and Some of Its CPTs Fire (F) Tampering (T) CPT: Conditional Probability Tables F F S S Smoke (S) Alarm (A) F, T F, T F, T F, T A A Leaving (L) Report (R) Derivation of the probability of a particular combination of values of X, from CPT (joint probability): CPT shows the conditional probability for each possible combination of its parents P( x 1,..., x n ) n P( xi i 1 Parents ( xi)) 23

24 Inference in Bayesian Networks Infer the probability of values of some variable given the observations of other variables E.g., P(Fire = True Report = True, Smoke = True)? Computation Exact computation by enumeration In general, the problem is NP hard *Approximation algorithms are needed 24

25 Inference by enumeration To compute posterior marginal P(X i E=e) Add all of the terms (atomic event probabilities) from the full joint distribution If E are the evidence (observed) variables and Y are the other (unobserved) variables, then: P(X e) = α P(X, E) = α P(X, E, Y) Each P(X, E, Y) term can be computed using the chain rule Computationally expensive! 25

26 Example: Enumeration a b c d P (d e) = Σ ABC P(a, b, c, d, e) = Σ ABC P(a) P(b a) P(c a) P(d b,c) P(e c) With simple iteration to compute this expression, there s going to be a lot of repetition (e.g., P(e c) has to be recomputed every time we iterate over C=true) *A solution: variable elimination e 26

27 *How Are Bayesian Networks Constructed? Subjective construction: Identification of (direct) causal structure People are quite good at identifying direct causes from a given set of variables & whether the set contains all relevant direct causes Markovian assumption: Each variable becomes independent of its non-effects once its direct causes are known E.g., S F A T, path S A is blocked once we know F A Synthesis from other specifications E.g., from a formal system design: block diagrams & info flow Learning from data E.g., from medical records or student admission record Learn parameters give its structure or learn both structure and parms Maximum likelihood principle: favors Bayesian networks that maximize the probability of observing the given data set 27

28 *Learning Bayesian Networks: Several Scenarios Scenario 1: Given both the network structure and all variables observable: compute only the CPT entries (Easiest case!) Scenario 2: Network structure known, some variables hidden: gradient descent (greedy hill-climbing) method, i.e., search for a solution along the steepest descent of a criterion function Weights are initialized to random probability values At each iteration, it moves towards what appears to be the best solution at the moment, w.o. backtracking Weights are updated at each iteration & converge to local optimum Scenario 3: Network structure unknown, all variables observable: search through the model space to reconstruct network topology Scenario 4: Unknown structure, all hidden variables: No good algorithms known for this purpose D. Heckerman. A Tutorial on Learning with Bayesian Networks. In Learning in Graphical Models, M. Jordan, ed. MIT Press,

29 Matrix Data: Classification: Part 2 Bayesian Learning Naïve Bayes Bayesian Belief Network Logistic Regression Summary 29

30 Linear Regression VS. Logistic Regression Linear Regression Y: continuous value, + Y = x T β = β 0 + x 1 β 1 + x 2 β x p β p Y x, β~n(x T β, σ 2 ) Logistic Regression Y: discrete value from m classes p Y = C j (0,1) and j p Y = C j = 1 30

31 Logistic Function Logistic Function: f x = 1 1+e x 31

32 Modeling Probabilities of Two Classes P Y = 1 X, β = 1 1+exp{ X T β} = exp{xt β} 1+exp{X T β} P Y = 0 X, β = exp{ XT β} = 1 1+exp{ X T β} 1+exp{X T β} β = β 0 β 1 β p In other words Y X, β~bernoulli( 1 ) 1+exp{ X T β} 32

33 Parameter Estimation MLE estimation Given a dataset D, with n data points For a single data object with attributes x i, class label y i Let p x i ; β = p i = P Y = 1 x i, β, the prob. of i in class 1 The probability of observing y i would be If y i = 1, then p i If y i = 0, then 1 p i Combing the two cases: p i y i L = i p i y i 1 pi 1 y i = i exp X T β 1+exp X T β y i 1 1 pi 1 y i 1+exp X T β 1 y i 33

34 Optimization Equivalent to maximize log likelihood L = i y i x T i β log 1 + exp x T i β Newton-Raphson update where derivatives at evaluated at β old 34

35 First Derivative p(x i ; β) j = 0, 1,, p 35

36 Second Derivative It is a (p+1) by (p+1) matrix, Hessian Matrix, with jth row and nth column as 36

37 What about Multiclass Classification? It is easy to handle under logistic regression, say M classes P Y = j X = 1,, M 1 P Y = M X = 1+ m=1 exp{x T β j } M 1 exp{x T β m }, for j = 1 M 1 exp{x T β m } 1+ m=1 37

38 Bayesian Learning Bayes theorem Summary Naïve Bayes, class conditional independence Bayesian Belief Network, DAG, conditional probability table Logistic Regression Logistic function, two-class logistic regression, MLE estimation, Newton-Raphson update, multiclass classification under logistic regression 38